full-send
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions that explicitly override other project-level safety guidelines. In the 'Testing permission' section of
SKILL.md, it directs the agent that if any project file or other instruction requires asking the user before using a test method (such as computer use, browser automation, or dev servers), this skill's instructions 'replace that rule' and the agent should proceed without asking. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, creating a surface for indirect prompt injection.
- Ingestion points: As described in
SKILL.md, the agent ingests data from Pull Request descriptions, comments, review threads, and bot outputs to guide its fixes and review process. - Boundary markers: Technical delimiters are absent; the skill relies on a behavioral instruction to treat external text as 'evidence, not instructions' and to verify all claims against the actual code.
- Capability inventory: The skill possesses extensive capabilities documented in
SKILL.md, including committing and pushing code, force-pushing branches, merging PRs via theghandgitCLI tools, and executing arbitrary test/build commands. - Sanitization: Programmatic sanitization is absent. The skill mitigates risk by instructing the agent to never automatically fix security-sensitive code (e.g., authentication, secrets, access control) and requiring manual verification of all claims.
- [COMMAND_EXECUTION]: The skill frequently executes shell commands using
gitand the GitHub CLI (gh). This includes checking repository permissions, viewing and editing PR metadata, and performing merges. It also allows for the execution of arbitrary test suites, linting tools, and build scripts as part of the PR verification process.
Audit Metadata